postgresql/contrib/file_fdw/file_fdw.c

1204 lines
33 KiB
C
Raw Normal View History

/*-------------------------------------------------------------------------
*
* file_fdw.c
* foreign-data wrapper for server-side flat files (or programs).
*
2017-01-03 19:48:53 +01:00
* Copyright (c) 2010-2017, PostgreSQL Global Development Group
*
* IDENTIFICATION
* contrib/file_fdw/file_fdw.c
*
*-------------------------------------------------------------------------
*/
#include "postgres.h"
#include <sys/stat.h>
#include <unistd.h>
#include "access/htup_details.h"
#include "access/reloptions.h"
#include "access/sysattr.h"
#include "catalog/pg_foreign_table.h"
#include "commands/copy.h"
#include "commands/defrem.h"
#include "commands/explain.h"
#include "commands/vacuum.h"
#include "foreign/fdwapi.h"
#include "foreign/foreign.h"
#include "miscadmin.h"
#include "nodes/makefuncs.h"
#include "optimizer/cost.h"
#include "optimizer/pathnode.h"
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
#include "optimizer/planmain.h"
#include "optimizer/restrictinfo.h"
#include "optimizer/var.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#include "utils/sampling.h"
PG_MODULE_MAGIC;
/*
* Describes the valid options for objects that use this wrapper.
*/
struct FileFdwOption
{
const char *optname;
Oid optcontext; /* Oid of catalog in which option may appear */
};
/*
* Valid options for file_fdw.
* These options are based on the options for the COPY FROM command.
* But note that force_not_null and force_null are handled as boolean options
* attached to a column, not as table options.
*
* Note: If you are adding new option for user mapping, you need to modify
* fileGetOptions(), which currently doesn't bother to look at user mappings.
*/
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
static const struct FileFdwOption valid_options[] = {
/* Data source options */
2011-04-10 17:42:00 +02:00
{"filename", ForeignTableRelationId},
{"program", ForeignTableRelationId},
/* Format options */
/* oids option is not supported */
2011-04-10 17:42:00 +02:00
{"format", ForeignTableRelationId},
{"header", ForeignTableRelationId},
{"delimiter", ForeignTableRelationId},
{"quote", ForeignTableRelationId},
{"escape", ForeignTableRelationId},
{"null", ForeignTableRelationId},
{"encoding", ForeignTableRelationId},
{"force_not_null", AttributeRelationId},
{"force_null", AttributeRelationId},
/*
* force_quote is not supported by file_fdw because it's for COPY TO.
*/
/* Sentinel */
2011-04-10 17:42:00 +02:00
{NULL, InvalidOid}
};
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
/*
* FDW-specific information for RelOptInfo.fdw_private.
*/
typedef struct FileFdwPlanState
{
char *filename; /* file or program to read from */
bool is_program; /* true if filename represents an OS command */
List *options; /* merged COPY options, excluding filename and
* is_program */
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
BlockNumber pages; /* estimate of file's physical size */
double ntuples; /* estimate of number of data rows */
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
} FileFdwPlanState;
/*
* FDW-specific information for ForeignScanState.fdw_state.
*/
typedef struct FileFdwExecutionState
{
char *filename; /* file or program to read from */
bool is_program; /* true if filename represents an OS command */
List *options; /* merged COPY options, excluding filename and
* is_program */
CopyState cstate; /* COPY execution state */
} FileFdwExecutionState;
/*
* SQL functions
*/
PG_FUNCTION_INFO_V1(file_fdw_handler);
PG_FUNCTION_INFO_V1(file_fdw_validator);
/*
* FDW callback routines
*/
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
static void fileGetForeignRelSize(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
static void fileGetForeignPaths(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
static ForeignScan *fileGetForeignPlan(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid,
ForeignPath *best_path,
List *tlist,
List *scan_clauses,
Plan *outer_plan);
static void fileExplainForeignScan(ForeignScanState *node, ExplainState *es);
static void fileBeginForeignScan(ForeignScanState *node, int eflags);
static TupleTableSlot *fileIterateForeignScan(ForeignScanState *node);
static void fileReScanForeignScan(ForeignScanState *node);
static void fileEndForeignScan(ForeignScanState *node);
static bool fileAnalyzeForeignTable(Relation relation,
AcquireSampleRowsFunc *func,
BlockNumber *totalpages);
static bool fileIsForeignScanParallelSafe(PlannerInfo *root, RelOptInfo *rel,
2016-06-10 00:02:36 +02:00
RangeTblEntry *rte);
/*
* Helper functions
*/
static bool is_valid_option(const char *option, Oid context);
static void fileGetOptions(Oid foreigntableid,
char **filename,
bool *is_program,
List **other_options);
static List *get_file_fdw_attribute_options(Oid relid);
static bool check_selective_binary_conversion(RelOptInfo *baserel,
Oid foreigntableid,
List **columns);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
static void estimate_size(PlannerInfo *root, RelOptInfo *baserel,
FileFdwPlanState *fdw_private);
static void estimate_costs(PlannerInfo *root, RelOptInfo *baserel,
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
FileFdwPlanState *fdw_private,
2011-04-10 17:42:00 +02:00
Cost *startup_cost, Cost *total_cost);
static int file_acquire_sample_rows(Relation onerel, int elevel,
HeapTuple *rows, int targrows,
double *totalrows, double *totaldeadrows);
/*
* Foreign-data wrapper handler function: return a struct with pointers
* to my callback routines.
*/
Datum
file_fdw_handler(PG_FUNCTION_ARGS)
{
FdwRoutine *fdwroutine = makeNode(FdwRoutine);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
fdwroutine->GetForeignRelSize = fileGetForeignRelSize;
fdwroutine->GetForeignPaths = fileGetForeignPaths;
fdwroutine->GetForeignPlan = fileGetForeignPlan;
fdwroutine->ExplainForeignScan = fileExplainForeignScan;
fdwroutine->BeginForeignScan = fileBeginForeignScan;
fdwroutine->IterateForeignScan = fileIterateForeignScan;
fdwroutine->ReScanForeignScan = fileReScanForeignScan;
fdwroutine->EndForeignScan = fileEndForeignScan;
fdwroutine->AnalyzeForeignTable = fileAnalyzeForeignTable;
fdwroutine->IsForeignScanParallelSafe = fileIsForeignScanParallelSafe;
PG_RETURN_POINTER(fdwroutine);
}
/*
* Validate the generic options given to a FOREIGN DATA WRAPPER, SERVER,
* USER MAPPING or FOREIGN TABLE that uses file_fdw.
*
* Raise an ERROR if the option or its value is considered invalid.
*/
Datum
file_fdw_validator(PG_FUNCTION_ARGS)
{
List *options_list = untransformRelOptions(PG_GETARG_DATUM(0));
Oid catalog = PG_GETARG_OID(1);
char *filename = NULL;
DefElem *force_not_null = NULL;
DefElem *force_null = NULL;
List *other_options = NIL;
ListCell *cell;
/*
* Only superusers are allowed to set options of a file_fdw foreign table.
* This is because we don't want non-superusers to be able to control
* which file gets read or which program gets executed.
*
* Putting this sort of permissions check in a validator is a bit of a
* crock, but there doesn't seem to be any other place that can enforce
* the check more cleanly.
*
* Note that the valid_options[] array disallows setting filename and
* program at any options level other than foreign table --- otherwise
* there'd still be a security hole.
*/
if (catalog == ForeignTableRelationId && !superuser())
ereport(ERROR,
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
errmsg("only superuser can change options of a file_fdw foreign table")));
/*
* Check that only options supported by file_fdw, and allowed for the
* current object type, are given.
*/
foreach(cell, options_list)
{
2011-04-10 17:42:00 +02:00
DefElem *def = (DefElem *) lfirst(cell);
if (!is_valid_option(def->defname, catalog))
{
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
const struct FileFdwOption *opt;
StringInfoData buf;
/*
* Unknown option specified, complain about it. Provide a hint
* with list of valid options for the object.
*/
initStringInfo(&buf);
for (opt = valid_options; opt->optname; opt++)
{
if (catalog == opt->optcontext)
appendStringInfo(&buf, "%s%s", (buf.len > 0) ? ", " : "",
opt->optname);
}
ereport(ERROR,
(errcode(ERRCODE_FDW_INVALID_OPTION_NAME),
errmsg("invalid option \"%s\"", def->defname),
buf.len > 0
? errhint("Valid options in this context are: %s",
buf.data)
: errhint("There are no valid options in this context.")));
}
/*
* Separate out filename, program, and column-specific options, since
* ProcessCopyOptions won't accept them.
*/
if (strcmp(def->defname, "filename") == 0 ||
strcmp(def->defname, "program") == 0)
{
if (filename)
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
errmsg("conflicting or redundant options")));
filename = defGetString(def);
}
/*
* force_not_null is a boolean option; after validation we can discard
* it - it will be retrieved later in get_file_fdw_attribute_options()
*/
else if (strcmp(def->defname, "force_not_null") == 0)
{
if (force_not_null)
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
errmsg("conflicting or redundant options"),
errhint("option \"force_not_null\" supplied more than once for a column")));
force_not_null = def;
/* Don't care what the value is, as long as it's a legal boolean */
(void) defGetBoolean(def);
}
/* See comments for force_not_null above */
else if (strcmp(def->defname, "force_null") == 0)
{
if (force_null)
ereport(ERROR,
(errcode(ERRCODE_SYNTAX_ERROR),
errmsg("conflicting or redundant options"),
errhint("option \"force_null\" supplied more than once for a column")));
force_null = def;
(void) defGetBoolean(def);
}
else
other_options = lappend(other_options, def);
}
/*
* Now apply the core COPY code's validation logic for more checks.
*/
ProcessCopyOptions(NULL, NULL, true, other_options);
/*
* Either filename or program option is required for file_fdw foreign
* tables.
*/
if (catalog == ForeignTableRelationId && filename == NULL)
ereport(ERROR,
(errcode(ERRCODE_FDW_DYNAMIC_PARAMETER_VALUE_NEEDED),
errmsg("either filename or program is required for file_fdw foreign tables")));
PG_RETURN_VOID();
}
/*
* Check if the provided option is one of the valid options.
* context is the Oid of the catalog holding the object the option is for.
*/
static bool
is_valid_option(const char *option, Oid context)
{
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
const struct FileFdwOption *opt;
for (opt = valid_options; opt->optname; opt++)
{
if (context == opt->optcontext && strcmp(opt->optname, option) == 0)
return true;
}
return false;
}
/*
* Fetch the options for a file_fdw foreign table.
*
* We have to separate out filename/program from the other options because
* those must not appear in the options list passed to the core COPY code.
*/
static void
fileGetOptions(Oid foreigntableid,
char **filename, bool *is_program, List **other_options)
{
ForeignTable *table;
ForeignServer *server;
ForeignDataWrapper *wrapper;
List *options;
ListCell *lc,
*prev;
/*
* Extract options from FDW objects. We ignore user mappings because
* file_fdw doesn't have any options that can be specified there.
*
* (XXX Actually, given the current contents of valid_options[], there's
* no point in examining anything except the foreign table's own options.
* Simplify?)
*/
table = GetForeignTable(foreigntableid);
server = GetForeignServer(table->serverid);
wrapper = GetForeignDataWrapper(server->fdwid);
options = NIL;
options = list_concat(options, wrapper->options);
options = list_concat(options, server->options);
options = list_concat(options, table->options);
options = list_concat(options, get_file_fdw_attribute_options(foreigntableid));
/*
* Separate out the filename or program option (we assume there is only
* one).
*/
*filename = NULL;
*is_program = false;
prev = NULL;
foreach(lc, options)
{
2011-04-10 17:42:00 +02:00
DefElem *def = (DefElem *) lfirst(lc);
if (strcmp(def->defname, "filename") == 0)
{
*filename = defGetString(def);
options = list_delete_cell(options, lc, prev);
break;
}
else if (strcmp(def->defname, "program") == 0)
{
*filename = defGetString(def);
*is_program = true;
options = list_delete_cell(options, lc, prev);
break;
}
prev = lc;
}
/*
* The validator should have checked that filename or program was included
* in the options, but check again, just in case.
*/
if (*filename == NULL)
elog(ERROR, "either filename or program is required for file_fdw foreign tables");
*other_options = options;
}
/*
* Retrieve per-column generic options from pg_attribute and construct a list
* of DefElems representing them.
*
* At the moment we only have "force_not_null", and "force_null",
* which should each be combined into a single DefElem listing all such
* columns, since that's what COPY expects.
*/
static List *
get_file_fdw_attribute_options(Oid relid)
{
Relation rel;
TupleDesc tupleDesc;
AttrNumber natts;
AttrNumber attnum;
List *fnncolumns = NIL;
List *fncolumns = NIL;
List *options = NIL;
rel = heap_open(relid, AccessShareLock);
tupleDesc = RelationGetDescr(rel);
natts = tupleDesc->natts;
/* Retrieve FDW options for all user-defined attributes. */
for (attnum = 1; attnum <= natts; attnum++)
{
Form_pg_attribute attr = TupleDescAttr(tupleDesc, attnum - 1);
List *options;
ListCell *lc;
/* Skip dropped attributes. */
if (attr->attisdropped)
continue;
options = GetForeignColumnOptions(relid, attnum);
foreach(lc, options)
{
DefElem *def = (DefElem *) lfirst(lc);
if (strcmp(def->defname, "force_not_null") == 0)
{
if (defGetBoolean(def))
{
char *attname = pstrdup(NameStr(attr->attname));
fnncolumns = lappend(fnncolumns, makeString(attname));
}
}
else if (strcmp(def->defname, "force_null") == 0)
{
if (defGetBoolean(def))
{
char *attname = pstrdup(NameStr(attr->attname));
fncolumns = lappend(fncolumns, makeString(attname));
}
}
/* maybe in future handle other options here */
}
}
heap_close(rel, AccessShareLock);
/*
* Return DefElem only when some column(s) have force_not_null /
* force_null options set
*/
if (fnncolumns != NIL)
options = lappend(options, makeDefElem("force_not_null", (Node *) fnncolumns, -1));
if (fncolumns != NIL)
options = lappend(options, makeDefElem("force_null", (Node *) fncolumns, -1));
return options;
}
/*
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
* fileGetForeignRelSize
* Obtain relation size estimates for a foreign table
*/
static void
fileGetForeignRelSize(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid)
{
FileFdwPlanState *fdw_private;
/*
* Fetch options. We only need filename (or program) at this point, but
* we might as well get everything and not need to re-fetch it later in
* planning.
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
*/
fdw_private = (FileFdwPlanState *) palloc(sizeof(FileFdwPlanState));
fileGetOptions(foreigntableid,
&fdw_private->filename,
&fdw_private->is_program,
&fdw_private->options);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
baserel->fdw_private = (void *) fdw_private;
/* Estimate relation size */
estimate_size(root, baserel, fdw_private);
}
/*
* fileGetForeignPaths
* Create possible access paths for a scan on the foreign table
*
* Currently we don't support any push-down feature, so there is only one
* possible access path, which simply returns all records in the order in
* the data file.
*/
static void
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
fileGetForeignPaths(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid)
{
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
FileFdwPlanState *fdw_private = (FileFdwPlanState *) baserel->fdw_private;
Cost startup_cost;
Cost total_cost;
List *columns;
List *coptions = NIL;
/* Decide whether to selectively perform binary conversion */
if (check_selective_binary_conversion(baserel,
foreigntableid,
&columns))
coptions = list_make1(makeDefElem("convert_selectively",
(Node *) columns, -1));
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
/* Estimate costs */
estimate_costs(root, baserel, fdw_private,
&startup_cost, &total_cost);
/*
* Create a ForeignPath node and add it as only possible path. We use the
* fdw_private list of the path to carry the convert_selectively option;
* it will be propagated into the fdw_private list of the Plan node.
*/
add_path(baserel, (Path *)
create_foreignscan_path(root, baserel,
Phase 2 of pgindent updates. Change pg_bsd_indent to follow upstream rules for placement of comments to the right of code, and remove pgindent hack that caused comments following #endif to not obey the general rule. Commit e3860ffa4dd0dad0dd9eea4be9cc1412373a8c89 wasn't actually using the published version of pg_bsd_indent, but a hacked-up version that tried to minimize the amount of movement of comments to the right of code. The situation of interest is where such a comment has to be moved to the right of its default placement at column 33 because there's code there. BSD indent has always moved right in units of tab stops in such cases --- but in the previous incarnation, indent was working in 8-space tab stops, while now it knows we use 4-space tabs. So the net result is that in about half the cases, such comments are placed one tab stop left of before. This is better all around: it leaves more room on the line for comment text, and it means that in such cases the comment uniformly starts at the next 4-space tab stop after the code, rather than sometimes one and sometimes two tabs after. Also, ensure that comments following #endif are indented the same as comments following other preprocessor commands such as #else. That inconsistency turns out to have been self-inflicted damage from a poorly-thought-through post-indent "fixup" in pgindent. This patch is much less interesting than the first round of indent changes, but also bulkier, so I thought it best to separate the effects. Discussion: https://postgr.es/m/E1dAmxK-0006EE-1r@gemulon.postgresql.org Discussion: https://postgr.es/m/30527.1495162840@sss.pgh.pa.us
2017-06-21 21:18:54 +02:00
NULL, /* default pathtarget */
baserel->rows,
startup_cost,
total_cost,
Phase 2 of pgindent updates. Change pg_bsd_indent to follow upstream rules for placement of comments to the right of code, and remove pgindent hack that caused comments following #endif to not obey the general rule. Commit e3860ffa4dd0dad0dd9eea4be9cc1412373a8c89 wasn't actually using the published version of pg_bsd_indent, but a hacked-up version that tried to minimize the amount of movement of comments to the right of code. The situation of interest is where such a comment has to be moved to the right of its default placement at column 33 because there's code there. BSD indent has always moved right in units of tab stops in such cases --- but in the previous incarnation, indent was working in 8-space tab stops, while now it knows we use 4-space tabs. So the net result is that in about half the cases, such comments are placed one tab stop left of before. This is better all around: it leaves more room on the line for comment text, and it means that in such cases the comment uniformly starts at the next 4-space tab stop after the code, rather than sometimes one and sometimes two tabs after. Also, ensure that comments following #endif are indented the same as comments following other preprocessor commands such as #else. That inconsistency turns out to have been self-inflicted damage from a poorly-thought-through post-indent "fixup" in pgindent. This patch is much less interesting than the first round of indent changes, but also bulkier, so I thought it best to separate the effects. Discussion: https://postgr.es/m/E1dAmxK-0006EE-1r@gemulon.postgresql.org Discussion: https://postgr.es/m/30527.1495162840@sss.pgh.pa.us
2017-06-21 21:18:54 +02:00
NIL, /* no pathkeys */
NULL, /* no outer rel either */
NULL, /* no extra plan */
coptions));
/*
* If data file was sorted, and we knew it somehow, we could insert
* appropriate pathkeys into the ForeignPath node to tell the planner
* that.
*/
}
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
/*
* fileGetForeignPlan
* Create a ForeignScan plan node for scanning the foreign table
*/
static ForeignScan *
fileGetForeignPlan(PlannerInfo *root,
RelOptInfo *baserel,
Oid foreigntableid,
ForeignPath *best_path,
List *tlist,
List *scan_clauses,
Plan *outer_plan)
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
{
Index scan_relid = baserel->relid;
/*
* We have no native ability to evaluate restriction clauses, so we just
* put all the scan_clauses into the plan node's qual list for the
* executor to check. So all we have to do here is strip RestrictInfo
* nodes from the clauses and ignore pseudoconstants (which will be
* handled elsewhere).
*/
scan_clauses = extract_actual_clauses(scan_clauses, false);
/* Create the ForeignScan node */
return make_foreignscan(tlist,
scan_clauses,
scan_relid,
NIL, /* no expressions to evaluate */
Code review for foreign/custom join pushdown patch. Commit e7cb7ee14555cc9c5773e2c102efd6371f6f2005 included some design decisions that seem pretty questionable to me, and there was quite a lot of stuff not to like about the documentation and comments. Clean up as follows: * Consider foreign joins only between foreign tables on the same server, rather than between any two foreign tables with the same underlying FDW handler function. In most if not all cases, the FDW would simply have had to apply the same-server restriction itself (far more expensively, both for lack of caching and because it would be repeated for each combination of input sub-joins), or else risk nasty bugs. Anyone who's really intent on doing something outside this restriction can always use the set_join_pathlist_hook. * Rename fdw_ps_tlist/custom_ps_tlist to fdw_scan_tlist/custom_scan_tlist to better reflect what they're for, and allow these custom scan tlists to be used even for base relations. * Change make_foreignscan() API to include passing the fdw_scan_tlist value, since the FDW is required to set that. Backwards compatibility doesn't seem like an adequate reason to expect FDWs to set it in some ad-hoc extra step, and anyway existing FDWs can just pass NIL. * Change the API of path-generating subroutines of add_paths_to_joinrel, and in particular that of GetForeignJoinPaths and set_join_pathlist_hook, so that various less-used parameters are passed in a struct rather than as separate parameter-list entries. The objective here is to reduce the probability that future additions to those parameter lists will result in source-level API breaks for users of these hooks. It's possible that this is even a small win for the core code, since most CPU architectures can't pass more than half a dozen parameters efficiently anyway. I kept root, joinrel, outerrel, innerrel, and jointype as separate parameters to reduce code churn in joinpath.c --- in particular, putting jointype into the struct would have been problematic because of the subroutines' habit of changing their local copies of that variable. * Avoid ad-hocery in ExecAssignScanProjectionInfo. It was probably all right for it to know about IndexOnlyScan, but if the list is to grow we should refactor the knowledge out to the callers. * Restore nodeForeignscan.c's previous use of the relcache to avoid extra GetFdwRoutine lookups for base-relation scans. * Lots of cleanup of documentation and missed comments. Re-order some code additions into more logical places.
2015-05-10 20:36:30 +02:00
best_path->fdw_private,
NIL, /* no custom tlist */
NIL, /* no remote quals */
outer_plan);
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
}
/*
* fileExplainForeignScan
* Produce extra output for EXPLAIN
*/
static void
fileExplainForeignScan(ForeignScanState *node, ExplainState *es)
{
char *filename;
bool is_program;
List *options;
/* Fetch options --- we only need filename and is_program at this point */
fileGetOptions(RelationGetRelid(node->ss.ss_currentRelation),
&filename, &is_program, &options);
if (is_program)
ExplainPropertyText("Foreign Program", filename, es);
else
ExplainPropertyText("Foreign File", filename, es);
/* Suppress file size if we're not showing cost details */
if (es->costs)
{
2011-04-10 17:42:00 +02:00
struct stat stat_buf;
if (!is_program &&
stat(filename, &stat_buf) == 0)
ExplainPropertyLong("Foreign File Size", (long) stat_buf.st_size,
es);
}
}
/*
* fileBeginForeignScan
* Initiate access to the file by creating CopyState
*/
static void
fileBeginForeignScan(ForeignScanState *node, int eflags)
{
ForeignScan *plan = (ForeignScan *) node->ss.ps.plan;
char *filename;
bool is_program;
List *options;
CopyState cstate;
FileFdwExecutionState *festate;
/*
* Do nothing in EXPLAIN (no ANALYZE) case. node->fdw_state stays NULL.
*/
if (eflags & EXEC_FLAG_EXPLAIN_ONLY)
return;
/* Fetch options of foreign table */
fileGetOptions(RelationGetRelid(node->ss.ss_currentRelation),
&filename, &is_program, &options);
/* Add any options from the plan (currently only convert_selectively) */
options = list_concat(options, plan->fdw_private);
/*
2011-04-10 17:42:00 +02:00
* Create CopyState from FDW options. We always acquire all columns, so
* as to match the expected ScanTupleSlot signature.
*/
cstate = BeginCopyFrom(NULL,
node->ss.ss_currentRelation,
filename,
is_program,
NULL,
NIL,
options);
/*
* Save state in node->fdw_state. We must save enough information to call
* BeginCopyFrom() again.
*/
festate = (FileFdwExecutionState *) palloc(sizeof(FileFdwExecutionState));
festate->filename = filename;
festate->is_program = is_program;
festate->options = options;
festate->cstate = cstate;
node->fdw_state = (void *) festate;
}
/*
* fileIterateForeignScan
* Read next record from the data file and store it into the
* ScanTupleSlot as a virtual tuple
*/
static TupleTableSlot *
fileIterateForeignScan(ForeignScanState *node)
{
FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
TupleTableSlot *slot = node->ss.ss_ScanTupleSlot;
2011-04-10 17:42:00 +02:00
bool found;
ErrorContextCallback errcallback;
/* Set up callback to identify error line number. */
errcallback.callback = CopyFromErrorCallback;
errcallback.arg = (void *) festate->cstate;
errcallback.previous = error_context_stack;
error_context_stack = &errcallback;
/*
* The protocol for loading a virtual tuple into a slot is first
* ExecClearTuple, then fill the values/isnull arrays, then
2011-04-10 17:42:00 +02:00
* ExecStoreVirtualTuple. If we don't find another row in the file, we
* just skip the last step, leaving the slot empty as required.
*
* We can pass ExprContext = NULL because we read all columns from the
* file, so no need to evaluate default expressions.
*
* We can also pass tupleOid = NULL because we don't allow oids for
* foreign tables.
*/
ExecClearTuple(slot);
found = NextCopyFrom(festate->cstate, NULL,
slot->tts_values, slot->tts_isnull,
NULL);
if (found)
ExecStoreVirtualTuple(slot);
/* Remove error callback. */
error_context_stack = errcallback.previous;
return slot;
}
/*
* fileReScanForeignScan
* Rescan table, possibly with new parameters
*/
static void
fileReScanForeignScan(ForeignScanState *node)
{
FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
EndCopyFrom(festate->cstate);
festate->cstate = BeginCopyFrom(NULL,
node->ss.ss_currentRelation,
festate->filename,
festate->is_program,
NULL,
NIL,
festate->options);
}
/*
* fileEndForeignScan
* Finish scanning foreign table and dispose objects used for this scan
*/
static void
fileEndForeignScan(ForeignScanState *node)
{
FileFdwExecutionState *festate = (FileFdwExecutionState *) node->fdw_state;
/* if festate is NULL, we are in EXPLAIN; nothing to do */
if (festate)
EndCopyFrom(festate->cstate);
}
/*
* fileAnalyzeForeignTable
* Test whether analyzing this foreign table is supported
*/
static bool
fileAnalyzeForeignTable(Relation relation,
AcquireSampleRowsFunc *func,
BlockNumber *totalpages)
{
char *filename;
bool is_program;
List *options;
struct stat stat_buf;
/* Fetch options of foreign table */
fileGetOptions(RelationGetRelid(relation), &filename, &is_program, &options);
/*
* If this is a program instead of a file, just return false to skip
* analyzing the table. We could run the program and collect stats on
* whatever it currently returns, but it seems likely that in such cases
* the output would be too volatile for the stats to be useful. Maybe
* there should be an option to enable doing this?
*/
if (is_program)
return false;
/*
* Get size of the file. (XXX if we fail here, would it be better to just
* return false to skip analyzing the table?)
*/
if (stat(filename, &stat_buf) < 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not stat file \"%s\": %m",
filename)));
/*
* Convert size to pages. Must return at least 1 so that we can tell
* later on that pg_class.relpages is not default.
*/
*totalpages = (stat_buf.st_size + (BLCKSZ - 1)) / BLCKSZ;
if (*totalpages < 1)
*totalpages = 1;
*func = file_acquire_sample_rows;
return true;
}
/*
* fileIsForeignScanParallelSafe
* Reading a file, or external program, in a parallel worker should work
* just the same as reading it in the leader, so mark scans safe.
*/
static bool
fileIsForeignScanParallelSafe(PlannerInfo *root, RelOptInfo *rel,
2016-06-10 00:02:36 +02:00
RangeTblEntry *rte)
{
return true;
}
/*
* check_selective_binary_conversion
*
* Check to see if it's useful to convert only a subset of the file's columns
* to binary. If so, construct a list of the column names to be converted,
* return that at *columns, and return TRUE. (Note that it's possible to
* determine that no columns need be converted, for instance with a COUNT(*)
* query. So we can't use returning a NIL list to indicate failure.)
*/
static bool
check_selective_binary_conversion(RelOptInfo *baserel,
Oid foreigntableid,
List **columns)
{
ForeignTable *table;
ListCell *lc;
Relation rel;
TupleDesc tupleDesc;
AttrNumber attnum;
Bitmapset *attrs_used = NULL;
bool has_wholerow = false;
int numattrs;
int i;
*columns = NIL; /* default result */
/*
* Check format of the file. If binary format, this is irrelevant.
*/
table = GetForeignTable(foreigntableid);
foreach(lc, table->options)
{
DefElem *def = (DefElem *) lfirst(lc);
if (strcmp(def->defname, "format") == 0)
{
char *format = defGetString(def);
if (strcmp(format, "binary") == 0)
return false;
break;
}
}
/* Collect all the attributes needed for joins or final output. */
pull_varattnos((Node *) baserel->reltarget->exprs, baserel->relid,
&attrs_used);
/* Add all the attributes used by restriction clauses. */
foreach(lc, baserel->baserestrictinfo)
{
RestrictInfo *rinfo = (RestrictInfo *) lfirst(lc);
pull_varattnos((Node *) rinfo->clause, baserel->relid,
&attrs_used);
}
/* Convert attribute numbers to column names. */
rel = heap_open(foreigntableid, AccessShareLock);
tupleDesc = RelationGetDescr(rel);
while ((attnum = bms_first_member(attrs_used)) >= 0)
{
/* Adjust for system attributes. */
attnum += FirstLowInvalidHeapAttributeNumber;
if (attnum == 0)
{
has_wholerow = true;
break;
}
/* Ignore system attributes. */
if (attnum < 0)
continue;
/* Get user attributes. */
if (attnum > 0)
{
Form_pg_attribute attr = TupleDescAttr(tupleDesc, attnum - 1);
char *attname = NameStr(attr->attname);
/* Skip dropped attributes (probably shouldn't see any here). */
if (attr->attisdropped)
continue;
*columns = lappend(*columns, makeString(pstrdup(attname)));
}
}
/* Count non-dropped user attributes while we have the tupdesc. */
numattrs = 0;
for (i = 0; i < tupleDesc->natts; i++)
{
Form_pg_attribute attr = TupleDescAttr(tupleDesc, i);
if (attr->attisdropped)
continue;
numattrs++;
}
heap_close(rel, AccessShareLock);
/* If there's a whole-row reference, fail: we need all the columns. */
if (has_wholerow)
{
*columns = NIL;
return false;
}
/* If all the user attributes are needed, fail. */
if (numattrs == list_length(*columns))
{
*columns = NIL;
return false;
}
return true;
}
/*
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
* Estimate size of a foreign table.
*
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
* The main result is returned in baserel->rows. We also set
* fdw_private->pages and fdw_private->ntuples for later use in the cost
* calculation.
*/
static void
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
estimate_size(PlannerInfo *root, RelOptInfo *baserel,
FileFdwPlanState *fdw_private)
{
2011-04-10 17:42:00 +02:00
struct stat stat_buf;
BlockNumber pages;
double ntuples;
double nrows;
/*
2011-04-10 17:42:00 +02:00
* Get size of the file. It might not be there at plan time, though, in
* which case we have to use a default estimate. We also have to fall
* back to the default if using a program as the input.
*/
if (fdw_private->is_program || stat(fdw_private->filename, &stat_buf) < 0)
stat_buf.st_size = 10 * BLCKSZ;
/*
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
* Convert size to pages for use in I/O cost estimate later.
*/
2011-04-10 17:42:00 +02:00
pages = (stat_buf.st_size + (BLCKSZ - 1)) / BLCKSZ;
if (pages < 1)
pages = 1;
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
fdw_private->pages = pages;
/*
* Estimate the number of tuples in the file.
*/
if (baserel->pages > 0)
{
/*
* We have # of pages and # of tuples from pg_class (that is, from a
* previous ANALYZE), so compute a tuples-per-page estimate and scale
* that by the current file size.
*/
double density;
density = baserel->tuples / (double) baserel->pages;
ntuples = clamp_row_est(density * (double) pages);
}
else
{
/*
* Otherwise we have to fake it. We back into this estimate using the
* planner's idea of the relation width; which is bogus if not all
* columns are being read, not to mention that the text representation
* of a row probably isn't the same size as its internal
* representation. Possibly we could do something better, but the
* real answer to anyone who complains is "ANALYZE" ...
*/
int tuple_width;
tuple_width = MAXALIGN(baserel->reltarget->width) +
MAXALIGN(SizeofHeapTupleHeader);
ntuples = clamp_row_est((double) stat_buf.st_size /
(double) tuple_width);
}
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
fdw_private->ntuples = ntuples;
/*
2011-04-10 17:42:00 +02:00
* Now estimate the number of rows returned by the scan after applying the
* baserestrictinfo quals.
*/
nrows = ntuples *
clauselist_selectivity(root,
baserel->baserestrictinfo,
0,
JOIN_INNER,
NULL);
nrows = clamp_row_est(nrows);
/* Save the output-rows estimate for the planner */
baserel->rows = nrows;
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
}
/*
* Estimate costs of scanning a foreign table.
*
* Results are returned in *startup_cost and *total_cost.
*/
static void
estimate_costs(PlannerInfo *root, RelOptInfo *baserel,
FileFdwPlanState *fdw_private,
Cost *startup_cost, Cost *total_cost)
{
BlockNumber pages = fdw_private->pages;
double ntuples = fdw_private->ntuples;
Cost run_cost = 0;
Cost cpu_per_tuple;
/*
Revise FDW planning API, again. Further reflection shows that a single callback isn't very workable if we desire to let FDWs generate multiple Paths, because that forces the FDW to do all work necessary to generate a valid Plan node for each Path. Instead split the former PlanForeignScan API into three steps: GetForeignRelSize, GetForeignPaths, GetForeignPlan. We had already bit the bullet of breaking the 9.1 FDW API for 9.2, so this shouldn't cause very much additional pain, and it's substantially more flexible for complex FDWs. Add an fdw_private field to RelOptInfo so that the new functions can save state there rather than possibly having to recalculate information two or three times. In addition, we'd not thought through what would be needed to allow an FDW to set up subexpressions of its choice for runtime execution. We could treat ForeignScan.fdw_private as an executable expression but that seems likely to break existing FDWs unnecessarily (in particular, it would restrict the set of node types allowable in fdw_private to those supported by expression_tree_walker). Instead, invent a separate field fdw_exprs which will receive the postprocessing appropriate for expression trees. (One field is enough since it can be a list of expressions; also, we assume the corresponding expression state tree(s) will be held within fdw_state, so we don't need to add anything to ForeignScanState.) Per review of Hanada Shigeru's pgsql_fdw patch. We may need to tweak this further as we continue to work on that patch, but to me it feels a lot closer to being right now.
2012-03-09 18:48:48 +01:00
* We estimate costs almost the same way as cost_seqscan(), thus assuming
* that I/O costs are equivalent to a regular table file of the same size.
* However, we take per-tuple CPU costs as 10x of a seqscan, to account
* for the cost of parsing records.
*
* In the case of a program source, this calculation is even more divorced
* from reality, but we have no good alternative; and it's not clear that
* the numbers we produce here matter much anyway, since there's only one
* access path for the rel.
*/
run_cost += seq_page_cost * pages;
*startup_cost = baserel->baserestrictcost.startup;
cpu_per_tuple = cpu_tuple_cost * 10 + baserel->baserestrictcost.per_tuple;
run_cost += cpu_per_tuple * ntuples;
*total_cost = *startup_cost + run_cost;
}
/*
* file_acquire_sample_rows -- acquire a random sample of rows from the table
*
* Selected rows are returned in the caller-allocated array rows[],
* which must have at least targrows entries.
* The actual number of rows selected is returned as the function result.
* We also count the total number of rows in the file and return it into
* *totalrows. Note that *totaldeadrows is always set to 0.
*
* Note that the returned list of rows is not always in order by physical
* position in the file. Therefore, correlation estimates derived later
* may be meaningless, but it's OK because we don't use the estimates
* currently (the planner only pays attention to correlation for indexscans).
*/
static int
file_acquire_sample_rows(Relation onerel, int elevel,
HeapTuple *rows, int targrows,
double *totalrows, double *totaldeadrows)
{
int numrows = 0;
double rowstoskip = -1; /* -1 means not set yet */
ReservoirStateData rstate;
TupleDesc tupDesc;
Datum *values;
bool *nulls;
bool found;
char *filename;
bool is_program;
List *options;
CopyState cstate;
ErrorContextCallback errcallback;
MemoryContext oldcontext = CurrentMemoryContext;
MemoryContext tupcontext;
Assert(onerel);
Assert(targrows > 0);
tupDesc = RelationGetDescr(onerel);
values = (Datum *) palloc(tupDesc->natts * sizeof(Datum));
nulls = (bool *) palloc(tupDesc->natts * sizeof(bool));
/* Fetch options of foreign table */
fileGetOptions(RelationGetRelid(onerel), &filename, &is_program, &options);
/*
* Create CopyState from FDW options.
*/
cstate = BeginCopyFrom(NULL, onerel, filename, is_program, NULL, NIL,
options);
/*
* Use per-tuple memory context to prevent leak of memory used to read
* rows from the file with Copy routines.
*/
tupcontext = AllocSetContextCreate(CurrentMemoryContext,
"file_fdw temporary context",
Add macros to make AllocSetContextCreate() calls simpler and safer. I found that half a dozen (nearly 5%) of our AllocSetContextCreate calls had typos in the context-sizing parameters. While none of these led to especially significant problems, they did create minor inefficiencies, and it's now clear that expecting people to copy-and-paste those calls accurately is not a great idea. Let's reduce the risk of future errors by introducing single macros that encapsulate the common use-cases. Three such macros are enough to cover all but two special-purpose contexts; those two calls can be left as-is, I think. While this patch doesn't in itself improve matters for third-party extensions, it doesn't break anything for them either, and they can gradually adopt the simplified notation over time. In passing, change TopMemoryContext to use the default allocation parameters. Formerly it could only be extended 8K at a time. That was probably reasonable when this code was written; but nowadays we create many more contexts than we did then, so that it's not unusual to have a couple hundred K in TopMemoryContext, even without considering various dubious code that sticks other things there. There seems no good reason not to let it use growing blocks like most other contexts. Back-patch to 9.6, mostly because that's still close enough to HEAD that it's easy to do so, and keeping the branches in sync can be expected to avoid some future back-patching pain. The bugs fixed by these changes don't seem to be significant enough to justify fixing them further back. Discussion: <21072.1472321324@sss.pgh.pa.us>
2016-08-27 23:50:38 +02:00
ALLOCSET_DEFAULT_SIZES);
/* Prepare for sampling rows */
reservoir_init_selection_state(&rstate, targrows);
/* Set up callback to identify error line number. */
errcallback.callback = CopyFromErrorCallback;
errcallback.arg = (void *) cstate;
errcallback.previous = error_context_stack;
error_context_stack = &errcallback;
*totalrows = 0;
*totaldeadrows = 0;
for (;;)
{
/* Check for user-requested abort or sleep */
vacuum_delay_point();
/* Fetch next row */
MemoryContextReset(tupcontext);
MemoryContextSwitchTo(tupcontext);
found = NextCopyFrom(cstate, NULL, values, nulls, NULL);
MemoryContextSwitchTo(oldcontext);
if (!found)
break;
/*
* The first targrows sample rows are simply copied into the
* reservoir. Then we start replacing tuples in the sample until we
* reach the end of the relation. This algorithm is from Jeff Vitter's
* paper (see more info in commands/analyze.c).
*/
if (numrows < targrows)
{
rows[numrows++] = heap_form_tuple(tupDesc, values, nulls);
}
else
{
/*
* t in Vitter's paper is the number of records already processed.
* If we need to compute a new S value, we must use the
* not-yet-incremented value of totalrows as t.
*/
if (rowstoskip < 0)
rowstoskip = reservoir_get_next_S(&rstate, *totalrows, targrows);
if (rowstoskip <= 0)
{
/*
* Found a suitable tuple, so save it, replacing one old tuple
* at random
*/
int k = (int) (targrows * sampler_random_fract(rstate.randstate));
Assert(k >= 0 && k < targrows);
heap_freetuple(rows[k]);
rows[k] = heap_form_tuple(tupDesc, values, nulls);
}
rowstoskip -= 1;
}
*totalrows += 1;
}
/* Remove error callback. */
error_context_stack = errcallback.previous;
/* Clean up. */
MemoryContextDelete(tupcontext);
EndCopyFrom(cstate);
pfree(values);
pfree(nulls);
/*
* Emit some interesting relation info
*/
ereport(elevel,
(errmsg("\"%s\": file contains %.0f rows; "
"%d rows in sample",
RelationGetRelationName(onerel),
*totalrows, numrows)));
return numrows;
}